64 results about "Visual methods" patented technology

The invention discloses a dynamic semantic vision SLAM (Simultaneous Localization and Mapping) method based on point and line feature adaptive weighting. The method comprises the following steps: firstly, obtaining potential dynamic region prior by combining target detection and image segmentation; then adaptive weighting is carried out on point and line features in two stages: in the first stage, initial weighting is carried out based on motion consistency check and dynamic levels, and feature matching and pose preliminary estimation are realized; in the second stage, weighting is further carried out on the key frame level through multi-view consistency verification, and weighted optimization and map maintenance are combined; and finally, loopback verification and global optimization are completed based on weighted features, and cumulative drift is effectively eliminated. According to the method, interference of dynamic and low-credibility features is suppressed through differential feature weighting, the adaptability to an intermittent moving object is improved through a double-stage weighting mechanism, and the stability and observability of a low-texture scene are enhanced through point-line combined modeling.

The invention belongs to the field of robot positioning and mapping algorithms, and provides a visual SLAM method and device for point-line feature fusion and a medium, and the method comprises the following steps: preprocessing an image, and obtaining point features and line features; matching the point features and the line features to obtain matched point features and matched line features; obtaining the initial pose of the camera through the matched point features and the matched line features, and constructing map points; constructing a three-dimensional map straight line through the matched line features; and optimizing the three-dimensional map points and the three-dimensional map straight lines. Complementary advantages are formed through extraction, matching, mapping and optimization processes of tight coupling point features and line features, and the fundamental problem of insufficient features in a weak texture environment is effectively solved.

The invention relates to the technical field of machine vision, in particular to a self-adaptive sorting system and method based on machine vision. Parcel image data on the sorting line is collected through an image collection device, and the image data is analyzed and recognized in combination with a machine vision method, so that real-time parcel information is obtained; based on the real-time package information, the real-time running state of the sorting line is analyzed and obtained, and a self-adaptive control decision of the sorting line is executed according to the real-time running state; and according to the self-adaptive control decision result, the sorting line is controlled to execute self-adaptive sorting treatment. According to the invention, intelligent and automatic operation of the sorting system can be realized.

The invention relates to the technical field of computer vision and intelligent monitoring, and particularly discloses a pavement ponding detection method based on image calibration and dynamic region segmentation, which comprises the following steps of: firstly, selecting a reference frame image in a ponding-free or slight ponding state from a road monitoring video, and manually marking a ponding region to generate a reference mask; the method is used as a unified detection reference standard. And then, frame extraction is performed on a to-be-detected video, pixel-level identification is performed on each frame of image by using a pre-trained semantic segmentation model, a water accumulation area of a current frame is automatically extracted, and a current water surface mask is generated. And dynamically judging the diffusion or fading trend of the accumulated water by calculating the difference value between the pixel area of the current frame accumulated water area and the pixel area of the reference mask accumulated water area. According to the method, the defects that a traditional sensor is high in deployment cost and complex in maintenance and an existing visual method is prone to being interfered and cannot be quantitatively changed are overcome, and low-cost, high-adaptability and quantifiable continuous automatic monitoring and early warning on the road surface ponding range is achieved.

The invention relates to the field of computer vision and mobile robot navigation, and discloses an object-level semantic vision SLAM method and system based on a mixed attention mechanism target detection network and an ellipsoid model. The system comprises five parallel thread modules, namely a semantic perception module, a visual tracking and repositioning module, a local mapping and fusion module, a loopback detection module and a global consistency optimization module. According to the method, local and global features of an image are extracted in parallel through a target detection network, and high-precision semantic observation is output; when the tracking is lost, the dual geometric constraint of the 2D internally tangent ellipsoid-3D object ellipsoid is utilized, and the P3P algorithm and the IoU cost function are matched to realize rapid relocation. In the mapping process, Gaussian-Wasserstein distance is introduced to measure semantic re-projection errors, and a joint objective function containing map point geometric errors and object semantic errors is constructed to carry out bundle adjustment. According to the method, the problem of feature extraction failure in motion blur and weak texture scenes is effectively solved, and the construction precision of the semantic map and the survivability of the system are remarkably improved.

The invention specifically discloses a target perception type dynamic visual SLAM method fusing point and line features, and relates to the technical field of visual SLAM. According to the method, a YOLOv8-seg model is combined with TensorRT reasoning to realize target detection and segmentation, and a DeepSORT algorithm is reconstructed to complete multi-target stable tracking; carrying out motion consistency check based on a luminosity consistency principle and an optical flow method; calculating a three-dimensional center of the target according to the detection frame, the mask depth and camera parameters to realize three-dimensional reconstruction of the target; point and line features are extracted and processed, and the line features are represented by Plcarbon coordinates; and constructing a nonlinear optimization model based on an ORB-SLAM3 framework, and jointly optimizing camera pose, map points and map line parameters. The method effectively improves the positioning precision and the system robustness in a dynamic scene, achieves the precise modeling of a static structure and the reliable generation of a dynamic target track, and is suitable for autonomous navigation scenes such as an unmanned aerial vehicle and a service robot.

The application discloses a visual SLAM method and system in a dynamic environment, and belongs to the technical field of visual images. The method comprises the following steps: acquiring image information and depth information acquired during movement of an RGB-D camera, so as to acquire a visual image and distance values of each pixel point; performing feature extraction on the visual image, so as to acquire a feature point set; performing target detection on the visual image, so as to acquire a potential dynamic region set; performing three-dimensional reconstruction on the feature points, so as to acquire three-dimensional space points; acquiring comparison results of the feature points and the three-dimensional space points and the potential dynamic region, so as to acquire dynamic suspicious points; updating a dynamic confidence degree, performing optimization demand verification on the dynamic suspicious points, so as to determine and suppress dynamic suppression points, and acquiring static feature points; performing pose optimization on the static feature points, so as to acquire stable static features, and performing three-dimensional reconstruction, so as to acquire a high-quality map. The method solves the problems of insufficient real dynamic suppression precision and easy misdeletion of static features in the prior art, and improves the construction quality of the map.

The invention discloses a belt deviation detection device and method, and relates to the technical field of visual method detection and deep learning, and the belt deviation detection method comprises the steps: capturing a belt operation image, carrying out the belt position region labeling, constructing an image semantic segmentation network model, inputting the belt operation image into the image semantic segmentation network model, and carrying out the image semantic segmentation network model. Acquiring a belt position area; an image semantic segmentation network model is combined with an ASPP multi-scale feature extraction network, the belt edge position is accurately recognized, and whether the image semantic segmentation network model achieves a set effect or not is judged when the average intersection-to-union ratio is smaller than or equal to an intersection-to-union ratio threshold value. Adding the belt operation images with the intersection-to-union ratio smaller than the intersection-to-union ratio threshold in the test set into the training set, and continuing to train the model; the model accuracy is continuously improved, all-weather continuous real-time monitoring of the belt running condition is achieved, the accuracy of belt deviation detection is improved, and the stability is high.

The invention discloses a rotating equipment vibration displacement detection method based on machine vision spatial filtering, and the method comprises the steps: collecting a rotating equipment operation video according to an industrial high-speed camera, and obtaining a two-dimensional image frame of rotating equipment; performing graying processing on the two-dimensional image frame to obtain a preprocessed two-dimensional image; selecting a rectangular region image with a preset size from a preset position of the first frame of preprocessed two-dimensional image; performing operation on rectangular area images selected at the same position of the preprocessed two-dimensional image on line frame by frame according to the frame sequence of the video to obtain vibration displacement increments in one or more angle directions as a continuous vibration displacement sequence in the corresponding angle direction; drawing an envelope spectrum according to the continuous vibration displacement sequence in one or more filtering directions; and calculating the theoretical fault characteristic frequency of the rolling bearing, and searching in the obtained envelope spectrum to complete fault type identification. According to the invention, a visual method can be combined with a signal processing technology, so that a barrier between visual information and vibration characteristics is broken.

The invention relates to the technical field of laser cutting processing, and discloses a laser cutting system based on image recognition excess material detection, which comprises the following steps: acquiring a reference polarization fingerprint of an excess material through a calibration module; then a detection module scans the surface of the excess material and collects a data stream containing position, total intensity of scattered light and light intensity signals of an original sector, and when the total intensity of the scattered light is sharply attenuated and the polarization mismatch degree between a scanning Mueller matrix and a reference polarization fingerprint, which is calculated in real time, exceeds a threshold value, the point is determined as a boundary point; and the contour generation module is used for preprocessing and sorting the point clouds and generating a three-dimensional contour model capable of accurately reflecting the surface fluctuation of the excess material. According to the method, the sensitivity of a traditional visual method to illumination and surface defects is effectively overcome, rapid and high-precision detection of the real three-dimensional contour of the excess material is achieved, and the recycling efficiency and the cutting quality of the excess material are remarkably improved.

The invention discloses a net cage fouling degree intelligent identification system and method based on electromagnetic-optical fusion perception, and belongs to the technical field of net cage fouling degree intelligent identification. The system comprises a mobile scanning platform, an integrated multi-mode sensing unit and a central processing unit; the integrated multi-mode sensing unit comprises an optical imaging module, an electromagnetic characteristic detection module and a positioning auxiliary unit. The method comprises the steps of cooperative scanning and multi-modal data acquisition, data fusion and contamination intelligent identification, accurate positioning and coordinate mapping, global contamination mapping, quantitative evaluation and cleaning decision support. According to the invention, through optical and electromagnetic multi-mode perception fusion, the identification precision of the types of the fouling organisms and the quantification accuracy of the coverage rate are greatly improved, and especially in a turbid water body, the limitation of a traditional single vision method is effectively made up; based on the global fouling distribution map generated by the fusion positioning technology, the operation mode change from blind traversal to accurate targeting is realized.

This invention discloses an adversarial robust visual Transformer method and system based on region hybrid experts, comprising: innovatively adding a region hybrid expert module to the original visual Transformer model. The region hybrid expert module includes a global expert, a central expert, and a region expert, combining image patch-to-region transformation and region-to-image patch transformation to achieve region semantic modeling, and introducing region adaptive adversarial loss and region alignment loss to suppress attention shift and region semantic drift caused by adversarial perturbations, thereby improving the classification accuracy and robustness of the image classification model under various adversarial attack scenarios.

The invention discloses a liquid level optical precision measurement method and system based on multi-depth reference characteristics, and belongs to the field of non-contact industrial measurement. The method comprises: taking an image including a luminescent liquid level, a container opening as a first reference feature, and at least one second reference feature located below the container and having a known size, from diagonally above the container; acquiring reference information such as the real size of each reference feature; based on the real size of each reference feature and the pixel size of each reference feature in the image, constructing a core effective scale conversion relation related to depth; and finally, the liquid level height is accurately solved by using the relation. According to the method, the concept of multi-depth reference is introduced, the problems of perspective distortion and scale uncertainty of a traditional single-reference-object vision method under the strabismus condition are fundamentally solved, and non-contact high-precision measurement of the liquid level is achieved.

The application discloses a visual SLAM method fusing point features and coplanar anchor points. The method comprises the following steps: extracting point features and line segment features in parallel for a current frame image; processing the line segment features extracted from the current frame image, screening and combining to generate coplanar anchor point features; matching the point features and the coplanar anchor point features extracted from the current frame image with features in a visual map established, to obtain a point feature matching pair set and a coplanar anchor point feature matching pair set; constructing a fusion re-projection error model according to the point feature matching pair set and the coplanar anchor point feature matching pair set, optimizing and solving the camera pose of the current frame image, to obtain an optimized camera pose; updating the visual map according to the optimized camera pose, and judging whether the current frame image is added as a new key frame. The application overcomes the deficiency in a low-texture scene, and avoids the explosive growth of calculation caused by complex multi-feature fusion.

A computer vision system, with at least one processor configured to: acquire, from a sports match video, a plurality of pieces of consecutive image data indicating a portion of the sports match video, the plurality of pieces of consecutive image data including a plurality of pieces of first consecutive image data that are consecutive and a plurality of pieces of second consecutive image data that are consecutive after the plurality of pieces of first consecutive image data; and execute an estimation, by using a machine learning model, of whether the portion is of a predetermined scene type.

The application provides a computer vision-based building deformation intelligent monitoring and data processing system, a monitoring setting module is a method summary process before monitoring, a monitoring position inclination correction module is a self position correction process in a monitoring process, and a data intelligent processing module is a data processing process in the monitoring process.The application is based on a computer vision method, the automatic aperture adjustment and the monitoring position inclination correction module can reduce the influence of self vibration in an algorithm, and high precision is achieved; the application combines the displacement correlation between monitoring points, automatically analyzes abnormal conditions of monitoring data, and completes intelligent correction and noise reduction processing of the data, establishes a platform to draw a displacement-time curve, visualizes structural displacement, and realizes intelligent deformation measurement and processing.

A Head-Mounted Display system together with associated techniques for performing accurate and automatic inside-out positional, user body and environment tracking for virtual or mixed reality are disclosed. The system uses computer vision methods and data fusion from multiple sensors to achieve real-time tracking. High frame rate and low latency is achieved by performing part of the processing on the HMD itself.

The invention discloses a visual SLAM method and system based on a dynamic environment, the system combines ORB feature points and EDLine feature lines for feature extraction, the robustness of the system to a weak texture environment is improved by using EDLine line features, and the problem of feature sparsification after dynamic elimination is solved; yOLOv12 is introduced to detect a dynamic target and generate a dynamic frame for the dynamic target, and the generated dynamic frame is utilized to remove dynamic features, so that dynamic environment interference is removed; in addition, the system also adopts a multi-thread asynchronous processing architecture to decouple and optimize a complex computing task and a core tracking task, and separate operation is carried out, so that the real-time performance of the system and the utilization rate of computing resources are improved. According to the method, the precision, the robustness and the real-time performance of the visual SLAM system in a complex dynamic environment are remarkably improved, real-time sensing, positioning and mapping tasks in the environment can be completed, and the method can be applied to the fields of automatic driving, navigation positioning, robots and the like.

The invention provides a visual SLAM method in a low-light environment based on image contrast optimization, and belongs to the technical field of image feature extraction. The invention aims to solve the problems of fuzzy details and difficult feature extraction of the dark part of the low-illumination image. Comprising the steps of S1, collecting a low-illumination image based on a visual sensor; s2, performing Gamma correction on the low-illumination image; s3, carrying out adaptive histogram equalization on the image after brightness distribution adjustment; and S4, Gaussian filtering is performed on the image after the overall gray histogram optimization, noise points in the image are suppressed, and a final enhanced image is obtained. According to the method, the extraction number of ORB feature points can be increased by 72.2% at most compared with that of an original image while relatively low operation time is guaranteed, the number of feature points extracted from noisy points by mistake can be remarkably reduced, and the robustness of positioning mapping can be effectively improved.

A computer vision system, with at least one processor configured to: acquire, from a sports match video, a plurality of pieces of consecutive image data indicating a portion of the sports match video, the plurality of pieces of consecutive image data including a plurality of pieces of first consecutive image data that are consecutive; and execute an estimation, by using a machine learning model, of whether the portion is of a predetermined scene type.

An AI-driven ophthalmic treatment system (20) includes a laser radiation source (48), a controller (44) integrated with an artificial intelligence processing unit, a neural network processing unit (NPU), an electronic control unit, and a storage module. The controller is configured to perform the following artificial intelligence-based operations: automatically identify and calibrate a plurality of local target regions (84) within an eyeball region (25) of a patient (22) by a deep learning model, and assign an appropriate predetermined laser energy to each target region, respectively, using a machine learning algorithm; controlling the radiation source to perform laser irradiation on at least the first target area; after the irradiation of the first target area is completed, recognizing the change of an eyeball structure through a real-time image analysis and computer vision method; and based on the identified structural change, dynamically inhibiting or adjusting predetermined energy output corresponding to the non-irradiated second target area through an AI decision model. Other implementation modes adopting artificial intelligence and machine learning technologies are also covered.

The application relates to the technical field of computer vision, in particular to a visual SLAM method based on 3D Gaussian spatter in a dynamic environment, a system and a readable storage medium. By performing instance segmentation on a scene image and extracting a prior dynamic mask, a potential dynamic mask is identified and fused to obtain a comprehensive dynamic mask; the comprehensive dynamic mask is optimized by using residual entropy and depth consistency, and then dynamic features are accurately removed; then, in order to balance the stability of pose tracking and the efficiency of mapping, a two-stage key frame screening strategy is constructed; on this basis, an improved 3D Gaussian spatter technology is further introduced to render and construct a scene graph; for each Gaussian ellipsoid, the volume and opacity weight are estimated, the multi-view contribution weight is calculated, the maximum weight is obtained by using a weighted average method based on the multi-view contribution weight, and pruning is performed. The application aims to solve the problem of how to reduce the computing power consumption in the 3D GS-SLAM scheme while ensuring the rendering quality.

The application provides a visual SLAM method based on motion segmentation, comprising the following steps: S1, for the collected RGB image, a motion segmentation method Rigidmask is used to detect potential dynamic objects and generate a dynamic object mask image, and a YOLO algorithm is used for instance segmentation to obtain an object mask image; S2, the corresponding relationship of the two mask images is matched to obtain a final mask image; S3, an ORB algorithm is used for feature extraction on the collected RGB image to obtain feature points; S4, the mask image is binarized, the feature points are matched with the obtained binarized mask image, whether the feature points are dynamic feature points is judged, if the feature points are dynamic feature points, the feature points are removed; finally, the remaining static feature points are obtained and used for subsequent pose matching and estimation. The application can remove the influence of dynamic objects on pose estimation in the visual SLAM process, reduce the interference of dynamic objects, and improve the positioning accuracy and the like.

The invention relates to the technical field of water environment monitoring, in particular to a water surface floating oil recognition sampling ship based on machine vision, which comprises a ship body, a navigation propulsion system, a visual recognition system, a sampling system and a control unit, the visual identification system comprises a double-spectrum visual module, and the double-spectrum visual module comprises a first visual sensor and a second visual sensor. The control unit generates a distribution diagram for representing the natural optical flicker intensity of the water surface based on the second waveband image data; based on the first wave band image data, identifying a suspected floating oil area; performing spatial position matching on the suspected floating oil area and an area with optical flicker intensity obviously lower than that of the surrounding water body in the distribution map, and determining the floating oil area according to a matching result; according to the invention, a dual-spectrum visual module is adopted and water surface flicker characteristic analysis based on a near-infrared band is combined as a criterion of floating oil identification, so that the problem of poor identification reliability of a traditional RGB visual method under complex light and shadow and water surface fluctuation is solved.

The application relates to the field of computer vision and mobile robot navigation, and discloses an object-level semantic visual SLAM method and system based on a hybrid attention mechanism target detection network and an ellipsoid model. The system comprises five parallel thread modules of semantic perception, visual tracking and repositioning, local mapping and fusion, loop detection and global consistency optimization. The application extracts local and global features of an image in parallel through a target detection network, and outputs high-precision semantic observations; when tracking is lost, a 2D inscribed ellipse-3D object ellipsoid dual geometric constraint is used to realize fast repositioning in cooperation with a P3P algorithm and an IoU cost function. In the mapping process, a Gaussian-Wasserstein distance is introduced to measure semantic re-projection errors, and a joint objective function containing map point geometric errors and object semantic errors is constructed to perform bundle adjustment. The application effectively solves the problem of feature extraction failure in motion blur and weak texture scenes, and significantly improves the construction accuracy of a semantic map and the survival ability of the system.

The invention relates to a dynamic environment vision SLAM (Simultaneous Localization and Mapping) method based on double-source dynamic mask tracking. The method comprises the following steps: inputting a color image into a lightweight instance segmentation model for reasoning, generating a lagged semantic dynamic mask of a prior dynamic object, and asynchronously transmitting the lagged semantic dynamic mask to a front-end tracking thread; in a front-end thread, key points are detected based on a lagged semantic dynamic mask, optical flow tracking is carried out to a current frame, an overlook depth histogram is generated through the current frame, and a semantic dynamic mask of the current frame is obtained through screening and reconstruction. The geometric dynamic mask is calculated after the region is eliminated, and after the geometric dynamic mask which is not detected in the previous frame is supplemented, the geometric dynamic mask and the semantic dynamic mask are combined and collected to obtain a final dynamic mask. And filtering dynamic features by using the final dynamic mask, completing pose estimation and back-end optimization based on static features, updating a TSDF map by only using a static part, removing a changed foreground structure during revisit, and outputting a synchronous positioning and mapping result. By adopting the method, the positioning precision and the mapping quality of the visual SLAM system of the small unmanned platform can be improved.

The invention relates to an SLAM (simultaneous localization and mapping) method based on a linear feature multi-constraint optimization model, belongs to the technical field of synchronous localization and map construction, and solves the problems that feature points are difficult to extract, errors are easy to occur in a matching process and the calculation cost is high in an existing visual SLAM method based on point features. The method comprises the following steps: acquiring continuous frame images shot by a camera; linear features of each frame of image are extracted and preprocessed, and matched linear features are obtained; the linear features obtained through matching are expressed as spatial linear features in a Pluecker coordinate form; mapping the spatial linear feature of the previous frame of image to the image plane of each frame to obtain a mapped linear feature; based on the mapping straight line features and the straight line features of all the frames, a multi-constraint optimization model is constructed and optimized, and pose variation is obtained; wherein the multi-constraint optimization model comprises a linear feature vertical direction residual error model and a linear feature horizontal direction residual error model. The accuracy of positioning and navigation is improved.

The invention discloses an end-to-end snapshot compression computer vision method based on a pseudo-random mask array. The method comprises the following steps of: performing single exposure on a video sequence subjected to space-time light intensity modulation by using a two-dimensional image sensor and a pseudo-random binary mask to obtain a corresponding two-dimensional (2D) compression measurement value; a video reconstruction is taken as a proxy task, a trained compression denoising auto-encoder model is utilized to extract potential spatial-temporal feature representations from the 2D compression measurement values, and the compression denoising auto-encoder model comprises a shared encoder and at least one task-specific decoder. The shared encoder is configured to extract the potential spatio-temporal feature representation from the 2D compressed measurements; and inputting the potential spatio-temporal feature representation into the decoder to obtain a dynamic result of the downstream computer vision task. According to the method, the image reconstruction complexity and the calculation cost are remarkably reduced, the excellent performance can still be kept especially in an extremely low illumination scene, and the privacy protection characteristic is achieved.

The patent discloses a lightweight litchi disease and pest detection method fusing double-path attention and feature compensation, and belongs to the technical field of image intelligent detection and plant disease control. In view of the problem of insufficient detection accuracy of small disease and pest targets in complex agricultural scenes, systematic improvements are made from two aspects of model architecture and feature optimization. On the one hand, a dynamic separable convolution module fusing attention mechanism is designed, and a double-path attention enhancement mechanism and a small target feature compensation mechanism are innovatively combined to optimize the extraction and fusion process of multi-scale features and significantly improve the recognition performance of the model for small disease and pest targets. On the other hand, the "attention mechanism and partial convolution cooperation" structure is creatively used in the Bottleneck part of the C2f module of the YOLOv8 model to enhance the sensitivity to small targets and the feature expression ability. The method effectively solves the problems of missed detection and high false detection rate of traditional visual methods for small target disease and pests in complex natural environments.

A system and method are provided for determining properties of a metal strip in a coating process. A pattern is projected onto or reflected by the strip, and images of the pattern are captured simultaneously from at least two different angles using visual sensors. These images are analyzed to determine properties of the strip, such as position, orientation, movement, and / or deformation. The system may use reference points formed by intersections of the pattern with the strip edges or interior features to construct a three-dimensional representation of the strip. Deformations may be identified using deflectometry and magnification effects. The determined properties can be used to dynamically adjust coating process parameters, such as air knife position, roll position and alignment, and edge baffle location, to improve coating uniformity and reduce defects.